Today's Fine-tuning & Training: Fastest-Growing Projects — May 04, 2026
Today's the Fine-tuning & Training space, we're seeing a surge in innovative approaches to optimize and enhance language models. From multimodal fine-tuning to task-aware servers, developers are pushing the boundaries of what's possible with AI training. With growth scores skyrocketing, it's clear that these tools are resonating with the community.
mattmireles/gemma-tuner-multimodal takes the top spot this week, boasting a staggering growth score of 42.81 and over 1,400 stars. This repository provides a fine-tuning framework for Gemma models using PyTorch and Metal Performance Shaders on Apple Silicon, allowing developers to train models with audio, images, and text inputs. Its massive growth can be attributed to the increasing demand for multimodal AI capabilities.
QingGo/engram-peft comes in second, with a growth score of 14.68 and 31 stars. This unofficial implementation of DeepSeek Engram enables users to inject high-capacity conditional memory into large language models (LLMs) via sparse retrieval PEFT without increasing inference FLOPs. Its growing popularity can be attributed to the need for more efficient and effective methods for fine-tuning LLMs.
UNfukashigi/Anima-LoRA-Factory is another notable repository, with a growth score of 9.93 and 25 stars. This user-friendly GUI tool allows users to train LoRAs (Low-Rank Adaptation) for the next-generation Anima diffusion models, making it easier to adapt these models to specific tasks. Its growth can be attributed to the increasing interest in diffusion-based AI models.
ZJU-OmniAI/GFT and semidark/kokoro-deutsch also made significant gains this week, with growth scores of 8.50 and 5.48 respectively. GFT proposes a new fine-tuning method using unbiased group advantages and dynamic coefficient rectification, while kokoro-deutsch provides a training recipe for fine-tuning Kokoro-82M on the German language. Both repositories are gaining traction due to their innovative approaches to fine-tuning.
Dynamis-Labs/spectralquant may have a lower growth score of 3.43, but its impressive 133 stars indicate significant interest in this repository's approach to breaking compression limits via spectral structure. This repository proposes a novel method for quantizing neural networks, which has piqued the interest of many developers.
Lastly, Mintzs/oogaboogalm and hlpun/Train-in-Silence round out our list, with growth scores of 2.52 and 1.11 respectively. oogaboogalm explores fine-tuning AI models to reduce token use, while Train-in-Silence provides a task-aware MCP server for LLM fine-tuning. Both repositories demonstrate the community's ongoing efforts to optimize and improve AI training.
Overall, Today's trends in Fine-tuning & Training highlight the community's drive to innovate and push the boundaries of what's possible with AI models.
mattmireles/gemma-tuner-multimodal takes the top spot this week, boasting a staggering growth score of 42.81 and over 1,400 stars. This repository provides a fine-tuning framework for Gemma models using PyTorch and Metal Performance Shaders on Apple Silicon, allowing developers to train models with audio, images, and text inputs. Its massive growth can be attributed to the increasing demand for multimodal AI capabilities.
QingGo/engram-peft comes in second, with a growth score of 14.68 and 31 stars. This unofficial implementation of DeepSeek Engram enables users to inject high-capacity conditional memory into large language models (LLMs) via sparse retrieval PEFT without increasing inference FLOPs. Its growing popularity can be attributed to the need for more efficient and effective methods for fine-tuning LLMs.
UNfukashigi/Anima-LoRA-Factory is another notable repository, with a growth score of 9.93 and 25 stars. This user-friendly GUI tool allows users to train LoRAs (Low-Rank Adaptation) for the next-generation Anima diffusion models, making it easier to adapt these models to specific tasks. Its growth can be attributed to the increasing interest in diffusion-based AI models.
ZJU-OmniAI/GFT and semidark/kokoro-deutsch also made significant gains this week, with growth scores of 8.50 and 5.48 respectively. GFT proposes a new fine-tuning method using unbiased group advantages and dynamic coefficient rectification, while kokoro-deutsch provides a training recipe for fine-tuning Kokoro-82M on the German language. Both repositories are gaining traction due to their innovative approaches to fine-tuning.
Dynamis-Labs/spectralquant may have a lower growth score of 3.43, but its impressive 133 stars indicate significant interest in this repository's approach to breaking compression limits via spectral structure. This repository proposes a novel method for quantizing neural networks, which has piqued the interest of many developers.
Lastly, Mintzs/oogaboogalm and hlpun/Train-in-Silence round out our list, with growth scores of 2.52 and 1.11 respectively. oogaboogalm explores fine-tuning AI models to reduce token use, while Train-in-Silence provides a task-aware MCP server for LLM fine-tuning. Both repositories demonstrate the community's ongoing efforts to optimize and improve AI training.
Overall, Today's trends in Fine-tuning & Training highlight the community's drive to innovate and push the boundaries of what's possible with AI models.